Selen writeup on Bunescu and Mooney

This is a review of bunescu_2006_subsequence_kernels_for_relation_extraction by user:Selen

They extract relations using a generalization of subsequent kernels which takes features that is a mixture of words and word classes. They apply their method to ACE dataset.

The assumptions they make is that the relations can be of three form , fore between , between and between after.

What I don't like:

How can you generalize the method to relations that is not of the form fore between between and between after? It could exist when there is a relation but the entities are missing in the sentence.

Their performance is not very impressive.

They don't compare their method to other methods with subsequent kernels.